LLM-Assisted Python Libraries Challenge Third-Party Dependencies
A new research initiative called zerodep investigates the potential of Python's standard library to serve as a substitute for widely-used third-party libraries, focusing on both accuracy and efficiency. This initiative features more than 40 single-file modules categorized into 12 groups, including serialization, networking, cryptography, agent protocols, and text processing. Each module is a reimplementation using only the standard library, developed with assistance from LLMs while adhering to strict guidelines: no external imports, single-file format, drop-in API compatibility, and essential correctness checks against the reference library. The research seeks to answer two main questions: when the standard library is adequate and if LLMs can produce correct and efficient code. The results may help alleviate dependency management challenges, minimize supply chain risks, and ease deployment in limited environments.
Key facts
- The zerodep project includes over 40 modules across 12 categories.
- Categories include serialization, networking, cryptography, agent protocols, and text processing.
- Each module is a single-file, stdlib-only reimplementation of a popular third-party library.
- Development was assisted by LLMs under strict constraints.
- Constraints include no external imports, single file, drop-in API compatibility, and mandatory correctness validation.
- The study empirically tests correctness and performance of stdlib-only implementations.
- The research addresses reducing dependency management overhead, supply chain risk, and deployment friction.
- The paper is available on arXiv with ID 2605.21405.
Entities
Institutions
- arXiv